Research And Application Of Traffic Flow Forecasting Based On Graph Convolutional Network

The traffic flow prediction problem mainly considers the historical time series characteristics of traffic flow to predict the traffic flow information of the current target node in the future.As a research hotspot in the field of intelligent transportation,it is of great significance for traffic guidance and path planning.Aiming at the shortcomings of the existing methods of analyzing and modeling the spatio-temporal information characteristics of traffic network,such as the unreasonable construction of road network topology,the singleness of data feature information,which leads to the failure to effectively extract the dynamic spatial dependence of urban road network,andto mine the potential traffic models,this paper proposes a new method based on predictive power score,PPS)is used to construct the dynamic topological structure,so as to realize the effective extraction of spatiotemporal features.Secondly,the dynamic topology structure and the physical distance based topology structure are effectively fused,and the gcgru(graph continuous network gated recurrent unit)spatio-temporal combination model is used to learn the spatio-temporal characteristics,and then used for traffic flow prediction.Experimental results show that the proposed topology construction method has higher prediction efficiency and accuracy.Aiming at the problems of high computational complexity and inability to deal with implicit abstract features in existing traffic flow prediction models,this paper improves a traffic flow prediction model framework stgct(spatial temporal graph continuous network transformer)based on representation learning,which mainly divides traffic data learning into two learning frameworks:spatial feature and temporal feature,The spatiotemporal feature data is represented in matrix space to avoid complex feature engineering.At the same time,learning into two learning frameworks:spatial feature and temporal feature,The spatiotemporal feature data is represented in matrix space to avoid complex feature engineering.At the same time,the attention mechanism can better extract the long-term dependence of time series and reduce the loss of prediction error.Experimental results show that the proposed algorithm can efficiently deal with abstract complex problems,and achieves good performance in many experimental evaluations.Based on the research of the above methods,combined with the application requirements,through the analysis and research of the traffic flow prediction task in many aspects,and using the existing traffic information sensor data,a set of traffic flow prediction system is designed and implemented,which can predict and evaluate the actual traffic flow.